How to find Local maxima in Kernel Density Estimation?

问题

I'm trying to make a filter (to remove outlier and noise) using kernel density estimators(KDE). I applied KDE in my 3D (d=3) data points and that gives me the probability density function (PDF) f(x). Now as we know local maxima of density estimation f(x) defined the centers of the clusters of data points. So my idea is to define the appropriate f(x) which will determine those clusters.

My question is how and what method will be better suited for this purpose of finding local maxima in f(x). If anyone can provide me some example code/ idea I will really appreciate it.

Here is the code to find the KDE which give f(x) in 3D data.

import numpy as np
from scipy import stats

data = np.array([[1, 4, 3], [2, .6, 1.2], [2, 1, 1.2],
         [2, 0.5, 1.4], [5, .5, 0], [0, 0, 0],
         [1, 4, 3], [5, .5, 0], [2, .5, 1.2]])
data = data.T 
kde = stats.gaussian_kde(data)
minima = data.T.min(axis=0)
maxima = data.T.max(axis=0)
space = [np.linspace(mini,maxi,20) for mini, maxi in zip(minima,maxima)]
grid = np.meshgrid(*space)
coords = np.vstack(map(np.ravel, grid))
#Evaluate the KD estimated pdf at each coordinate
density = kde(coords)

回答1:

You will want to use an algorithm called Mean Shift. Its a clustering algorithm that works by finding the modes (aka maxima of f(x)) of the KDE. Note that the bandwidth set for your KDE will impact the number of modes and their locations. Since you are using python, there is an implementation in scikit-learn.

回答2:

Here is a short function that demonstrates how you could estimate the maxima. Note: the higher the number of no_samples the more accurate the maxima.

from scipy.stats import gaussian_kde
import numpy as np

    def estimate_maxima(data):

      kde = gaussian_kde(data)

      no_samples = 10

      samples = np.linspace(0, 10, no_samples)

      probs = kde.evaluate(samples)

      maxima_index = probs.argmax()

      maxima = samples[maxima_index]

      return maxima

来源：https://stackoverflow.com/questions/31198020/how-to-find-local-maxima-in-kernel-density-estimation

标签

python

machine-learning

cluster-analysis

kernel-density